See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260249223

Traffic Signal Control with Connected Vehicles

Article  in  Transportation Research Record Journal of the Transportation Research Board · December 2013
DOI: 10.3141/2381-08

CITATIONS READS
64 1,190

3 authors, including:

Noah Joseph Goodall Brian L. Smith

Virginia Transportation Research Council University of Virginia
15 PUBLICATIONS   293 CITATIONS    160 PUBLICATIONS   2,515 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Connected and Automated Vehicle View project

Preparing for Automated Vehicles in Virginia View project

All content following this page was uploaded by B. Brian Park on 04 September 2015.

The user has requested enhancement of the downloaded file.

 Goodall, Smith, and Park 1

1 TRAFFIC SIGNAL CONTROL WITH CONNECTED VEHICLES

2
3 Noah J. Goodall*, P.E.
4 Research Scientist
5 Virginia Center for Transportation Innovation and Research
6 530 Edgemont Road
7 Charlottesville, Virginia 22903
8 Phone: (434) 293-1905 Fax: (434) 434-1990
9 Email: [email protected]
10 *Corresponding Author
11
12 Brian L. Smith, Ph.D, P.E.
13 Professor and Chair
14 Department of Civil and Environmental Engineering
15 University of Virginia
16 PO Box 400742
17 351 McCormick Road
18 Charlottesville, Virginia 22904-4742
19 Phone: (434) 243-8585 Fax: (434) 982-2951
20 Email: [email protected]
21
22 Byungkyu (Brian) Park, Ph.D.
23 Associate Professor
24 Department of Civil and Environmental Engineering
25 University of Virginia
26 PO Box 400742
27 351 McCormick Road
28 Charlottesville, Virginia 22904-4742
29 Phone: (434) 924-6347 Fax: (434) 982-2951
30 Email: [email protected]
31
32 Word Count: 4,868 words
33 Graphs, Tables, and Figures: 9 at 250 words per
34 Total Count: 7,118 words
35
36 Submitted to:
37 AHB15 – Intelligent Transportation Systems
38
39
40 for Presentation and Publication at TRB 92nd Annual Meeting
41
42 November 15, 2012
43

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 2

1 ABSTRACT
2 The operation of traffic signals is currently limited by the data available from traditional point
3 sensors. Point detectors, often in-ground inductive loop sensors, can provide only limited vehicle
4 information at a fixed location. The most advanced adaptive control strategies are often not
5 implemented in the field due to their operational complexity and high resolution detection
6 requirements. However, a new initiative known as connected vehicles would allow for the
7 wireless transmission of vehicles’ positions, headings, and speeds to be used by the traffic
8 controller. A new traffic control algorithm, the predictive microscopic simulation algorithm
9 (PMSA), was developed in this research to utilize these new, more robust data. The
10 decentralized, fully adaptive traffic control algorithm uses a rolling horizon strategy, where the
11 phasing is chosen to optimize an objective function over a 15-second period in the future. The
12 objective function uses either delay-only, or a combination of delay, stops, and decelerations. To
13 measure the objective function, the algorithm uses a microscopic simulation driven by present
14 vehicle positions, headings, and speeds. Unlike most adaptive control strategies, the algorithm is
15 relatively simple, does not require point detectors or signal-to-signal communication, and is
16 completely responsive to immediate vehicle demands. To ensure drivers’ privacy, the algorithm
17 stores no memory of individual or aggregate vehicle locations. Results from simulation show that
18 the algorithm maintains or improves performance compared to a state-of-practice coordinated-
19 actuated timing plan optimized by Synchro at low- and mid-level volumes, but performance
20 worsens during saturated and oversaturated conditions. Testing also showed improved
21 performance during periods unexpected high demand and the ability to automatically respond to
22 year-to-year growth without retiming.
23
24 Key words: Connected vehicles, vehicle-to-infrastructure communication, IntelliDrive, traffic
25 signals, data communication, adaptive control, complex systems
26

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 3

1 INTRODUCTION
2 Traffic signals, when operated efficiently, can enable safe and efficient movement of vehicles
3 through an intersection and minimize delays in a corridor. However, most signal timing plans in
4 use must ignore or make assumptions about many aspects of traffic conditions. Fixed time
5 control, where a signal system uses a static and repeating sequence of phases and durations
6 designed to serve a certain time period, has no way to detect vehicles and therefore relies on the
7 expected approach volumes from manual traffic counts. Actuated timing plans use point
8 detectors to modify a fixed timing plan, by occasionally skipping a phase if no vehicle is present,
9 or shortening a phase when vehicles are not being served. Some adaptive timing plans attempt to
10 adjust to slow or systematic changes in volumes. Split Cycle Offset Optimisation Technique
11 (SCOOT) (1) and Signal Coordination and Timing (SCAT) (2) are two prominent examples.
12 However, both are restricted in that they only alter a cyclic timing plan.
13 Other timing plans do allow acyclic operation, but differ in their approach. The most
14 common technique is to use the concept of rolling horizon, where a traffic control algorithm
15 attempts to optimize an objective function such as delay over a short period into the future,
16 generally one or two cycle lengths. The objective function is optimized by estimating the
17 positions of vehicles during the horizon over several possible phasings. Some examples of point
18 detector-based rolling horizon strategies are ALLONS-D (3), UTOPIA (4), PRODYN (5), OPAC
19 (6), and RHODES (7). However, because these strategies are point detector-based, they are
20 forced to make several estimations, including a vehicle’s precise positions after passing a
21 detector, queue length, and vehicle speeds. Additionally, these adaptive control strategies are not
22 widely implemented in the United States, due primarily to their operational complexity
23 (engineers typically require 4-6 months to understand the systems) and maintenance demands
24 (8).

25 Connected Vehicle Wireless Detection Systems

26 A new initiative to allow wireless communication between vehicles and the transportation
27 infrastructure, referred to here as “connected vehicles,” may have broad implications for how
28 traffic signal control will operate in the future. Instead of relying on point detectors (such as
29 inductive loops or video detection systems) that sense only presence at fixed locations, signal
30 systems would be able to use data from in-vehicle sensors transmitted wirelessly from equipped
31 vehicles to the signal controller. Traffic signal control logic would have access to many measures
32 that were previously estimated or unknown such as vehicle speeds, positions, arrival rates, rates
33 of acceleration and deceleration, queue lengths, and stopped time.
34 A clear definition of the types of data and communications used by connected vehicles is
35 found in the Society of Automotive Engineers (SAE) J2735 Dedicated Short Range
36 Communications (DSRC) Message Set Dictionary (9). This standard defines vehicle-to-vehicle
37 and vehicle-to-infrastructure communications using DSRC, the medium-range communications
38 channels dedicated for vehicle use by the Federal Communications Commission (FCC) in 1999.
39 For safety applications, each vehicle transmits a Basic Safety Message (BSM) that transmits its
40 temporary ID, location, speed, heading, lateral and longitudinal acceleration, brake system status,
41 and vehicle size to surrounding vehicles and the infrastructure. By “listening” to these messages,
42 a signal controller could gain a more comprehensive understanding of the movements of nearby
43 vehicles than with loop and video detection.
44
45 Traffic Signal Control Using Individual Vehicle Locations

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 4

1 Several traffic signal timing plans have been proposed which utilize some form of wireless
2 communication between vehicles and the signal controller. Premier and Friedrich (10) proposed
3 a rolling horizon algorithm using vehicle-to-infrastructure communications based on the IEEE
4 802.11 standard. The algorithm sought to minimize queue lengths by optimizing phases in five-
5 second intervals over a 20-second horizon using the techniques of dynamic programming and
6 complete enumeration on an acyclic, decentralized system.
7 Datesh et al. (11) proposed an algorithm which uses vehicle clustering to apply a
8 sophisticated form of actuated control. The acyclic timing plan assigns the next phase to the first
9 group of queued vehicles to surpass a predetermined cumulative waiting time threshold. The
10 phase is extended to allow the next platoon to pass, which are identified using k-means clustering
11 based vehicles’ speeds and locations.
12 Lee (12) proposed the Cumulative Travel Time Responsive (CTR) Real-Time
13 Intersection Control Algorithm. This algorithm uses connected vehicles to determine the amount
14 of time that a vehicle has spent traveling to the intersection from within 300 meters or the nearest
15 intersection, whichever is closer. The travel time includes the time that the vehicle is in motion,
16 as well as its stopped time at the intersection, if any. The algorithm then sums the travel times for
17 each combination of movements (i.e. NEMA phases 2 & 6, or 4 & 8). The phasing with the
18 highest combined travel time is selected as the next phase, with a minimum green time of five
19 seconds. To supplement the travel time figures obtained at less than 100% market penetration, a
20 Kalman filtering technique was used to estimate actual cumulative travel times based on a
21 prediction of future travel times and the measurement of sampled vehicles.
22 He et al. (13) proposed the platoon-based arterial multi-modal signal control with online
23 data (PAMSCOD) algorithm, which used mixed integer linear programming to determine
24 phasing and timings every 30 seconds for four cycles in the future, based on predicted vehicle
25 platoon sizes and locations. PAMSCOD was able to improve vehicle and bus delay at saturation
26 rates greater than 0.8, but often experienced higher delays at saturation rates less than 0.6. The
27 saturation rate was calculated using Synchro’s intersection capacity utilization metric (14).
28 To date, no research has investigated utilizing microscopic simulation as a tool to
29 estimate future conditions in a rolling horizon algorithm in a connected vehicle environment
30 without re-identifying vehicles. Unlike previous connected vehicle signal control algorithms that
31 required at least short-term tracking of vehicle locations (e.g. to measure platoon movements or
32 waiting time), this research proposed the first signal control algorithm to use wireless vehicle
33 locations without re-identification or short-term tracking of vehicles. Furthermore, no other
34 research has investigated multi-objective optimization over the short-term horizon and its affect
35 on delay in the long-term in a connected vehicle environment.
36
37 PROPOSED TRAFFIC SIGNAL CONTROL ALGORITHM DESCRIPTION
38 The traffic signal control algorithm proposed here, called the predictive microscopic simulation
39 algorithm or PMSA, has the following three objectives:
40 1) to match or significantly improve the performance of a state-of-practice actuated-
41 coordinated system;
42 2) to respond to real-time demands only, thereby eliminating the need for manual timing
43 plan updates to adjust for traffic growth or fluctuations; and
44 3) to never re-identify, track, or store any records of individual or aggregate vehicle
45 movements for any length of time, thereby protecting driver privacy.
46

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 5

1 To accomplish these objectives, the PMSA uses a rolling horizon approach, where the
2 traffic signal controller attempts to minimize an objective function over a short time period in the
3 future. Although many detector-based traffic signal control strategies use rolling horizon, they
4 require complicated algorithms to estimate vehicle arrivals (15) and delay (3). They also require
5 reliable and highly accurate detection, generally in the form of loop detectors both at the
6 intersection and upstream of each approach. The failure of one or more detectors could be
7 catastrophic for the rolling horizon approach.
8 The PMSA, uses microscopic traffic simulation to simulate vehicles over the horizon
9 period, and calculates the objective function delay directly from the vehicle’s simulated
10 behavior. For the purposes of this description, an intersection’s movement is defined as a single
11 controlled vehicle path, e.g. westbound left, whereas a phase is defined as two non-contradictory
12 movements, e.g. westbound left and eastbound left. When the algorithm recalculates the signal’s
13 phase, it first collects a snapshot of the position, heading, and speed of every equipped vehicle
14 within 300 meters of the intersection (at 45 mi/hr, the speed of this corridor, a vehicle travels
15 exactly this distance during the 15-second horizon). This information is then used to populate a
16 model of the intersection, as shown in Figure 1.
17

Model Intersection

Field Intersection

18
19
FIGURE 1 The PMSA populating a model of the intersection with the positions and speed of the
20
equipped vehicles from the “real world” field intersection.
21
22 Once the model has been populated with the new vehicles, the vehicles are simulated
23 fifteen seconds into the future. Because the turn lanes in the test network were between 75 and
24 300 meters in length, the turning movement of many vehicles can be assumed based on their

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 6

1 current lane. For vehicle's upstream of the turning lane, it was assumed that 50% of those in the
2 lane nearest a turning lane would use the turning lane. This is repeated once for each possible
3 new phase configuration, as well as for the possibility of maintaining the current phasing. Four-
4 second amber phases and two-second red phases are simulated as well. The phase with the
5 optimal objective function over the fifteen second horizon is selected as the next phase.
6 The new phase’s green time is determined from the horizon simulation as the time
7 required to clear all simulated vehicles from a single movement. This time is bound with a
8 minimum of 5 seconds, and a maximum time before recalculation of 15 seconds.
9 To ensure smooth operation of the signal, several restrictions are put in place. Because
10 the algorithm is acyclic and allows phase skipping, each movement has a maximum red time of
11 120 seconds. This was considered reasonable, as the Synchro-recommended timing plan for the
12 corridor was 120 seconds. Also, to take advantage of the queue detection capabilities of
13 connected vehicles, the algorithm will not allow queues to block a turning lane or through lane.
14 When a vehicle is detected within 40 feet of blocking a movement, the vehicle’s movement is
15 given priority at the next phase recalculation. The PMSA’s decision process is shown in Figure
16 2.
17

Phase yes Vehicles no no

120s of
Start timed out? blocking a red?
movement?

no yes yes
Movement Movement
Next
priority = 1, priority = 2,
time step
“medium” “high”

Calculate Phase with Green time =

Simulate phases
objective lowest objective min{simulated time
with highest
function for each function chosen to clear, horizon
combined priority
phase selected length}

18
19 FIGURE 2 Predictive microscopic simulation algorithm's logic flow chart.
20
21 The algorithm operates completely without loop or video detection, with no knowledge
22 of expected demand or memory of past demand, and is completely decentralized. There is no
23 communication with any other signal on the corridor, either ad hoc or through synchronized
24 timing. The algorithm was designed to be compatible with the SAE J2735 standard for DSRC
25 communications. It requires only the information required in the Basic Safety Message no more
26 than once per second, whereas the message is sent ten times per second according to the
27 standard. Further, the algorithm is able protect driver privacy by clearing any vehicle data
28 seconds after it is recorded. Specifically, the algorithm does not store any vehicle location data,
29 neither aggregated volumes nor individual vehicle trajectories, once the next phase has been
30 determined.

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 7

1
2 SIMULATION TESTING AND RESULTS
3 To simulate the connected vehicle environment, the microscopic simulation software package
4 VISSIM was used, as it allows users to easily access individual vehicle information via a COM
5 interface, and allows a second “future” simulation to run parallel to the primary simulation. For
6 this study, a program was written in the C# programming language using VISSIM’s COM
7 interface to extract individual vehicle characteristics such as speed and position no more than
8 once per second.
9 The test network is a calibrated model of four intersections along Route 50 in Chantilly,
10 Virginia, shown in Figure 3. Vehicle volumes and turning movements were collected in 2003
11 between 3:00PM and 4:00PM on weekdays (16). Pedestrian movements, which were very low at
12 these intersections, were eliminated for the purpose of this analysis, as the minimum pedestrian
13 crossing time often exceeded 60 seconds, well beyond the algorithm’s 15-second horizon.
14

790 ft
North

1280 ft

1320 ft

15
16 FIGURE 3 Map of the test segment, a 4-signal stretch of US-50 in Chantilly, Virginia.
17
18 Vehicle volumes were converted to approximate intersection saturation rates using
19 Synchro’s intersection capacity utilization (ICU) metric (14), and measured at an average of
20 0.75. To test the sensitivity of the algorithm to various equipped vehicle penetration rates, the
21 algorithm was tested at 10%, 25%, 50%, and 100% vehicle participation, using total delay over
22 the horizon as the sole element of the objective function. Each scenario was evaluated for 30
23 minutes after 400 seconds of simulation initialization (17). Each scenario was tested ten times at
24 different random seeds, and all produced statistically similar results with a 95% confidence level
25 (18).

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 8

1 Off-line signal system optimization tools such as Synchro (19) and TRANSYT-7F (20)
2 are often used in practice to develop timing plans. In this research, Synchro was used to develop
3 an optimized coordinated-actuated timing plan with a 120 cycle length as a base case for
4 comparison with the PMSA. Synchro’s recommended timing plans were programmed into and
5 tested in the VISSIM network.
6
7
8 Single Variable Objective Function
9 The algorithm attempts to optimize some objective function over a 15 second interval. Initial
10 testing focused on a single variable, the cumulative vehicle delay. Although decentralized, the
11 algorithm produced coordinated flow, as evidenced by the green bands in Figure 4 showing
12 eastbound traffic.
13

14
15 FIGURE 4 Eastbound signal system coordination of the PMSA
16
17 TABLE 1 Performance of PMSA at Various Equipped Vehicle Penetration Rates
Equipped
Vehicle Average Stopped
Penetration Delay Speed Delay Stops
Rate Method (s/veh) (mi/hr) (s/veh)
Coordinated-
- 54.20 27.98 30.90 6807
Actuated
10% PMSA 66.11* 26.01* 34.79* 11431*
25% PMSA 58.09* 27.32* 27.15* 10233*

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 9

50% PMSA 52.26* 28.35* 22.98* 9254*

100% PMSA 51.06* 28.61* 25.46* 7417*
* n = 10, p < 0.05

1 The results of the testing are shown in Table 1. Improvements in delay and speed are only
2 experienced at penetration rates of 50% and higher. There are fewer stops at higher penetration
3 rates, but always more stops compared to a coordinated-actuated system. Stopped delay
4 improves at 25% penetration and higher, with a 34% improvement experienced with 50% of
5 vehicles participating. It should be noted that these improvements are experience with neither
6 any assumed knowledge of historical demand volumes, nor of any coordination or
7 communication with neighboring signals.
8 According to Federal Highway Administration (FHWA) estimates, 25% of congestion is
9 caused by incidents (21). The PMSA, because it requires no knowledge of historical traffic
10 demands, has the advantage of responding to unexpected demands due to incidents with minimal
11 transition time compared to a time-of-day plan. To evaluate the PMSA’s ability to handle large
12 unexpected variations in flow, a simulation was run where volumes entering the mainline
13 heading east increased by 30%. This represents a realistic scenario for vehicles rerouting to avoid
14 an incident on a parallel freeway or arterial. The PMSA, operating with 100% equipped vehicle
15 penetration rate, is able to respond instantly to the increased demand, with no outside input from
16 operators or communication with roadside infrastructure or nearby signals. With the unexpected
17 increase, the PMSA is able to reduce delay by 25.6%, increase speed by 12.8%, reduce stopped
18 delay by 27.4%, and reduce stops by 35.3%.
19
20 TABLE 2 Performance of PMSA during an unexpected 30% increase in volumes from the west.
Average Stopped
Delay Speed Delay Stops
Method (s/veh) (mi/hr) (s/veh)
Coordinated-Actuated 78.47 24.42 39.61 14260
PMSA 58.33* 27.54* 28.76* 9233*
Percent Change -25.6% 12.8% -27.4% -35.3%
* n = 10, p < 0.05

21 Another common cause of congestion is poor signal timing, estimated by FHWA to be

22 responsible for 5% of all congestion (21). Many transportation agencies lack the resources to
23 update signal timing plans every three years as recommended (22). Annual traffic growth, if not
24 addressed, can quickly overwhelm a timing plan and lead to poor performance. The PMSA,
25 because it responds only to immediate traffic demand, can accommodate annual volume
26 increases without adjustments. To test the PMSA’s ability, the algorithm was tested at 100%
27 market penetration against a coordinated-actuated timing plan that was optimized for the much
28 lower ten-year-old volumes ten years in the past, assuming a 3% annual growth rate for all
29 volumes. This equates to a 34% volume increase in all directions, with no change to the timing
30 plan. The results of this analysis are shown in Table 3. The PMSA showed significant benefits
31 across all metrics.

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 10

1
2 TABLE 3 Performance of PMSA on a Network with 3% Annual Volume Growth against a 10-
3 year Old Coordinated-Actuated Timing Plan
Average Stopped
Delay Speed Delay Stops
Method (s/veh) (mi/hr) (s/veh)
Coordinated-Actuated 69.85 25.50 39.86 9486
PMSA 51.06* 28.61* 25.46* 7417*
Percent Change -26.9% 12.2% -36.1% -21.8%
* n = 10, p < 0.05

4 The PMSA was additionally tested at uniformly varying volumes. Volumes at all
5 intersection were uniformly multiplied by 0.5, 0.75, 1, and 1.25, which by using Synchro’s
6 intersection capacity utilization metric as a surrogate, translated into approximate average
7 saturation rates of 0.45, 0.60, 0.75, and 0.90 respectively. Different optimized coordinated-
8 actuated timing plans were developed using Synchro for each of the new volume scenarios,
9 while the PMSA was tested unaltered regardless of the volumes, and assuming 100% equipped
10 vehicle penetration rate. The results are shown in Table 4. The PMSA either improves on or has
11 no significant effect on delay, average speed, or stopped delay below 0.90 saturation rate. There
12 are fewer stops at a saturation rate of 0.45, no difference at 0.60, and increased stops at 0.75 and
13 0.90. These results are similar to Lämmer and Helbing's decentralized adaptive systems, where
14 performance degrades at near-saturated conditions (23).
15
16 TABLE 4 Performance of PMSA at various saturation rates

Average Stopped
Saturation Delay Speed Delay Stops
Rate** Method (s/veh) (mi/hr) (s/veh)
0.45 Coordinated-Actuated 41.79 30.51 24.49 2818
PMSA 39.95 30.72 24.41 2612*
0.60 Coordinated-Actuated 51.75 28.62 30.63 4868
PMSA 50.22 28.72 29.89 4825
0.75 Coordinated-Actuated 54.20 27.98 30.90 6807
PMSA 51.06* 28.61* 25.46* 7417*
0.90 Coordinated-Actuated 69.81 25.61 41.03 12640
PMSA 77.87* 24.32* 40.13 19102*
*n = 10, p < 0.05
**Approximate average saturation rate based on Synchro’s ICU metric
17

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 11

1 Multivariable Objective Function

2 The previous analysis used vehicle delay as the sole variable in the objective function. Several
3 other variables were tested as well if factors other than delay could improve performance of the
4 algorithm. The objective function is described in Equation 1:
5
6 (1) f = αd + βa + γs
7 where α, β, and γ are factors, d is delay per second per vehicle, a is negative acceleration per
8 second per vehicle, and s is stops per vehicle. Delay and stops are commonly used measures of
9 effectiveness for signal timing. Negative acceleration was selected because of its relationship
10 with emissions; while positive acceleration is correlated with emissions, it is an unrealistic metric
11 in practice as it discourages phase changes under all circumstances. Negative acceleration is a
12 leading indicator of positive acceleration, and is therefore an acceptable surrogate measure. The
13 variables are defined in Equations 2-4:
 n t



∑∑ d ij 

14 (2) d = max d max , i=1 j =1 
 ntd max 
 

 n t

 ∑∑ max{− aij ,0}
 
15 (3) a = max amax , i =1 j =1 
 ntamax 
 

 n t



∑∑ sij 

16 (4) s = max smax , i=1 j =1 
 nsmax 
 
17 where the terms dmax, amax, and smax are terms used to cap and normalize the observed values, and
18 are set at 1 s/s/veh, 2 m/s3/veh, and 3 stops/veh, respectively.
19 A range of factors were tested, between 0 and 1 at intervals of 0.1, and ensuring that the
20 sum of α, β, and γ were equivalent to 1. The 30 combinations with the lowest average delay are
21 shown in Table 5, with the delay-only single variable objective function as a baseline. Results
22 that are significantly different from the delay-only results are marked in bold (p < 0.05). All
23 scenarios assume an equipped vehicle penetration rate of 100%.
24
25

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 12

1 TABLE 5 Results from Using a Multivariable Objective Function

Delay Acceleration Average Average Average
Factor Factor Stops Factor Delay Decleration Stops
α β γ (s/veh) (m/s2/veh/sec)
1 0 0 51.06 0.665 7417
0.8 0.0 0.2 50.68 0.668 7504
0.9 0.0 0.1 50.82 0.661 7440
0.6 0.0 0.4 51.79 0.674 7579
0.9 0.1 0.0 51.88 0.645* 7557
0.7 0.0 0.3 52.77 0.669 7670
0.8 0.1 0.1 52.86 0.650 7725
0.5 0.0 0.5 53.20* 0.668 7944
0.8 0.2 0.0 53.20* 0.647* 7756
0.6 0.1 0.3 53.92* 0.656 7631
0.7 0.1 0.2 54.00* 0.660 7821
0.3 0.1 0.6 54.19* 0.662 7824
0.3 0.0 0.7 54.33* 0.669 7861
0.6 0.2 0.2 54.34* 0.635* 7613
0.5 0.1 0.4 54.38* 0.653 7814
0.6 0.3 0.1 54.72* 0.634* 7621
0.7 0.2 0.1 54.94* 0.645* 7579
0.4 0.0 0.6 54.98* 0.662 7833
0.7 0.3 0.0 55.45* 0.645* 7873
0.5 0.2 0.3 55.53* 0.639* 7870
0.6 0.4 0.0 56.12* 0.639* 7934
0.4 0.2 0.4 56.25* 0.646 8004
0.2 0.0 0.8 56.25* 0.672 8064
0.5 0.3 0.2 56.38* 0.628* 7966
0.4 0.1 0.5 56.42* 0.655 8236*
0.2 0.1 0.7 56.60* 0.666 8167*
0.1 0.1 0.8 56.62* 0.659 8075*
0.4 0.4 0.2 56.83* 0.632* 7898
0.3 0.2 0.5 56.90* 0.651* 8116*
0.1 0.0 0.9 57.11* 0.671 8391
*n = 5, p < 0.05
2
3 Different objective functions were unable to significantly improve on the delay-only function,
4 either in average delay or stops. A high acceleration factor in particular produces poor
5 performance when compared to a delay-only function.
6
7 SUMMARY
8 A rolling horizon traffic signal control algorithm called PMSA is presented in this paper.
9 The algorithm uses individual vehicle locations, headings, and speeds to predict an objective
10 function over a 15-second future horizon using microscopic simulation. The algorithm does not
11 use any data from point detectors or any historical demands, nor does it require any

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 13

1 communication between signals. An important feature of the algorithm is that it only uses
2 instantaneous vehicle data, and does not re-identify or track vehicles in any way to protect
3 privacy.
4 Microscopic simulation shows that PMSA, using delay as the sole variable in the
5 objective function, is able to significantly improve or have no effect on the performance of
6 coordinated-actuated systems in several scenarios, specifically at low and medium demand
7 saturation, and with greater than 50% penetration rate of equipped vehicles. The algorithm shows
8 much greater improvements during unexpected demands for which the baseline Synchro
9 coordinated actuated timing plan was not optimized, particularly in a simulated incident and with
10 annual traffic volume increases where the timing plan is not updated. Different horizon objective
11 functions using delay, deceleration, and stops as variables were unable to improve the
12 performance of a delay-only function.
13 Future work will involve improving the performance of the algorithm at low connected
14 vehicle penetration rates. Recent research suggests that the behavior of a few connected vehicles
15 can estimate positions of unequipped vehicles in real-time on freeways (24), (25) and delayed on
16 arterials (26) These techniques may be adapted for signal control, where they can provide real-
17 time estimates of individual vehicle locations, thereby artificially augmenting the equipped
18 penetration rate. The algorithm will also be compared to existing adaptive control algorithms.
19

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 14

1 REFERENCES
2 (1) Hunt, P. B., R. D. Bretherton, D. I. Robertson, and M. C. Royal. SCOOT Online Traffic
3 Signal Optimsation Technique. Traffic Engineering and Control, Vol. 23, 1982. pp. 190–
4 192.
5 (2) Sims, A. and K. W. Dobinson. The Sydney Coordinated Adaptive Traffic (SCAT) System
6 Philosophy and Benefits. IEEE Transportation Vehicle Technology, Vol. 29, No. 2, May
7 1980, pp. 130–137.
8 (3) Porche, I. and S. Lafortune. Adaptive Look-ahead Optimization of Traffic Signals. Journal
9 of Intelligent Transportation Systems, Vol. 4, No. 3, 1999, pp. 209–254.
10 (4) Mauro, V. and C. Di Taranto. UTOPIA. Presented at the the Sixth IFAC/IFIP/IFORS
11 Symposium on Control and Communications in Transportation, Paris, France, 1990.
12 (5) Henry, J.J., J. L. Farges, and J. Tufal. The PRODYN Real Time Traffic Algorithm.
13 Presented at the 4th IFAC/IFIP/IFORS International Conference on Control in
14 Transportation Systems, Baden, 1983, pp. 307–311.
15 (6) Gartner, N. H. OPAC: A Demand-Responsive Strategy for Traffic Signal Control. In
16 Transportation Research Record: Journal of the Transportation Research Board, No. 906,
17 Transportation Research Board of the National Academies, Washington, D.C., 1983, pp.
18 75–81.
19 (7) Mirchandani, P. A. Real-time Traffic Signal Control System: Architecture, Algorithms,
20 and Analysis. Transportation Research Part C: Emerging Technologies, Vol. 9, No. 6,
21 December 2001, pp. 415–432.
22 (8) Stevanovic, A. Adaptive Traffic Control Systems: Domestic and Foreign State of Practice.
23 NCHRP Synthesis 403. National Research Council, Transportation Research Board,
24 National Cooperative Highway Research Program, American Association of State
25 Highway and Transportation Officials, and Federal Highway Administration, 2010.
26 (9) Society of Automotive Engineers. Dedicated Short Range Communications (DSRC)
27 Message Set Dictionary. Standard SAE J2735, Nov. 2009.
28 (10) Priemer, C. and B. Friedrich. A Decentralized Adaptive Traffic Signal Control Using V2I
29 Communication Data. Presented at 12th International IEEE Conference on Intelligent
30 Transportation Systems, St. Louis, MO, 2009, pp. 1–6.
31 (11) Datesh, J., W. T. Scherer, and B. L. Smith. Using K-Means Clustering to Improve Traffic
32 Signal Efficacy in an IntelliDrive Environment. Presented at 2011 IEEE Forum on
33 Integrated and Sustainable Transportation Systems, Vienna, Austria, June 2011.
34 (12) Lee, J. Assessing the Potential Benefits of IntelliDrive-based Intersection Control
35 Algorithms. Dissertation, University of Virginia, Charlottesville, VA, 2010.
36 (13) He, Q., K. L. Head, and J. Ding. PAMSCOD: Platoon-based Arterial Multi-modal Signal
37 Control with Online Data. Transportation Research Part C: Emerging Technologies, Vol.
38 20, February 2012, pp. 164–184.
39 (14) Husch, D. and J. Albeck. Intersection Capacity Utilization. Trafficware, 2003.
40 www.trafficware.com/assets/pdfs/ICU2003.pdf. Accessed July 2, 2012.
41 (15) Head, K. L. Event-Based Short-Term Traffic Flow Prediction Model. In Transportation
42 Research Record: Journal of the Transportation Research Board, No. 1510, Transportation
43 Research Board of the National Academies, Washington, D.C., 1995, pp. 45-52.
44 (16) Park, B., and J. D. Schneeberger. Microscopic Simulation Model Calibration and
45 Validation: Case Study of VISSIM Simulation Model for a Coordinated Actuated Signal
46 System. In Transportation Research Record: Journal of the Transportation Research

TRB 2013 Annual Meeting Paper revised from original submittal.

 Goodall, Smith, and Park 15

1 Board, No. 1856, Transportation Research Board of the National Academies, Washington,
2 D.C., 2003, pp. 185–192.
3 (17) Dowling, R., A. Skabardonis, and V. Alexiadis. Traffic Analysis Toolbox Volume III:
4 Guidelines for Applying Traffic Microsimulation Software. Publication FHWA-HRT-04-
5 040. FHWA, U.S. Department of Transportation, June 2004.
6 (18) Law, A. Simulation Modeling and Analysis, 4th Edition. McGraw-Hill, Dubuque , Iowa,
7 2007.
8 (19) Synchro Studio 8.0 User’s Guide. Trafficware, June 2011.
9 (20) TRANSYT-7F User’s Guide. TRANSYT-7F Version 11.3, McTrans, University of Florida,
10 Gainesville, Florida, 2011.
11 (21) Focus on Congestion Relief | Describing the Congestion Problem. FHWA, 2004.
12 www.fhwa.dot.gov/congestion/describing_problem.htm. Accessed July 20, 2012.
13 (22) National Transportation Operations Coalition. National Traffic Signal Report Card.
14 Institute of Transportation Engineers, 2012.
15 (23) Lämmer, S. and D. Helbing. Self-control of Traffic Lights and Vehicle Flows in Urban
16 Road Networks. Journal of Statistical Mechanics: Theory and Experiment, Vol. 4, April
17 2008.
18 (24) Herrera, J. C., and A. M. Bayen. Incorporation of Lagrangian Measurements in Freeway
19 Traffic State Estimation. Transportation Research Part B: Methodological, Vol. 44, No. 4,
20 May 2010, pp. 460–481.
21 (25) Goodall, N. J., B. L. Smith, and B. Park. Microscopic Estimation of Freeway Vehicle
22 Positions Using Mobile Sensors. Presented at 90th Annual Meeting of the Transportation
23 Research Board, Washington D.C., 2012.
24 (26) Sun Z., and X. Ban. Vehicle Trajectory Reconstruction for Signalized Intersections Using
25 Variational Formulation of Kinematic Waves. Presented at 89th Annual Meeting of the
26 Transportation Research Board, Washington D.C., 2011.
27

TRB 2013 Annual Meeting Paper revised from original submittal.

View publication stats